karalets

Overall loop for training a deep net for molecules here

29

We have two desiderata: 1. We want to be able to learn a network which regresses to measurements given a structure as input. 2. We may want to pretrain parts...

Graph generative models

15

Graph generative models are important for the tasks we have been describing. The core idea is to posit a model which defines some distribution over graphs ```P(G)```, for instance via...

What Metrics do we want to have for training/testing here?

5

Initially, we can have things like log-likelihood in order to just be able to get some reasonable quantitative thing. Over time, however, we may want to have more informative metrics...

How to define a model and a loss

29

https://github.com/choderalab/pinot/blob/e1d326043f2157f8572518a609c981cb0f444941/pinot/app/gaussian_variational.py#L16 Here, you define a loss by handwriting some math. This is doable, but I have some problems with it. First, the loss is basically the negative loglikelihood of the...

I just bumped across this paper here which builds probabilistic graph models and does some form of RL for property prediction and maximization: https://jcheminf.biomedcentral.com/articles/10.1186/s13321-019-0396-x I encourage everyone to have a...